wals.pro AI 4 weclapp

Each tool has a sharply distinct role, and descriptions explicitly cross-reference sibling tools (e.g. execute_api states that using it when search_entities/get_entity cover the need is a tool-selection error). Read vs. preview vs. execute phases, and per-domain preview_* tools (write_entity, write_comment, write_supply_source, stock_operation, upload_document, entity_action, escalate_to_support), are cleanly separated with no overlapping purpose.

All 25 tools use consistent snake_case with a predictable verb_noun pattern (search_entities, get_entity, preview_write_entity, execute_approved, read_activity_log, find_reconciliation_candidates, etc.). The only nominal deviation is tenant_health_check, but it is still idiomatic snake_case and readable.

25 tools is at the heavy end for a single server, and the seven preview_* variants plus multiple read/search tools push it past the comfortably scoped 3-15 range. Each tool does cover genuinely distinct functionality in a large ERP domain, but the surface is large enough to feel heavyweight for agents to route across.

The surface covers the full lifecycle: search/get/aggregate reads, create/update/delete previews via preview_write_entity, workflow and domain actions, document upload/download, comments, stock operations, reference data, schema discovery, activity log, settings, reconciliation matching, and tenant health. No obvious domain gaps or dead ends remain for the stated ERP-integration purpose.